Clayton Macleod wrote:
re: memory reporting, even the "working set" value you get from the
'performance monitor' doesn't really necessarily tell you how much RAM
the process is actually using just for itself, because this value
includes any shared memory, not just private memory. I'm honestly
sorry that I can't recall which MS guy said it, or exactly how he
phrased it, but the gist of it was that you really cannot tell exactly
how much RAM/pagefile is being used by processes since some of the
data being reported to you includes memory that is shared between
processes, and some of the values only relate to virtual memory space
allocated.
The most common cause of misrepresentation is the fact that Windows
pre-pages most data to prevent massive delays in freeing physical ram
when necessary. Was it Wang that was talking about that? Might have
been, and IIRC it was discussed during beta 2.
I just remember reading this during one of the windows
betas in MS's beta newsgroups, possibly the initial release of XP. I
imagine it was in the 'performance' group and came from one of the
guys that deals with memory management. Yeah, yeah, anecdotal. But
if you look through their various memory management/monitoring
articles you get basically the same information. Most of the memory
usage values reported are vritual memory stats, not real/physical
memory, and even the value for the working set is 'dirtied' by shared
memory. You can get a good enough idea from all the various reported
values, yeah. But you can't get exact figures. Hell, they even
changed task manager's title from mem usage to PF usage, and for good
reason.
Just because taskmgr doesn't report accurately does not mean the system
cannot account for all memory. Wang would be most upset (and probably
out of a job) if this was the case. Complete enumeration of these values
is costly however, which is why it's unecessary for taskmgr. It's almost
more important to have the values provided anyway.
Setting that option *does* reduce paging. I'm not disputing that.
phew.
I
only disputed that it is going to change the paging behaviour of
applications.
It does, because all changes to paging rates will change paging rates of
other applications too. Call it what you will, starvation, pre-tension,
or any of the other terms that people have tried to use to coin the
factor of side-effects within dynamic caching algorithms, but it's
princliple is the same. If there is something that needs to be regularly
accessed but is not regularly scheduled it can cause failures (well,
fail is too strong a word, but you know.) in the algorithms which can be
reduced by changing their run-time settings. This is simply what happens
in this scenario. Never underestimate how active the kernel and driver
pages are!
Clearly it will not, since applications and their
memory don't fall under the umbrella of this setting, which only deals
with the executive (NTExecutive I believe is the proper name) and the
things which belong to it. i.e. kernel and drivers.
Please don't try to tell me that the executive memory space is
unimportant, I know that you already know this isn't true.
On 8/15/05, James Tucker <[EMAIL PROTECTED]> wrote:
The discussions on MSDN are not important here, as the process scheduler
was made to report accurately for SP2(XP) and SP1(2k3). Amoung other
things this is a requirement for DEP internals. Most of these docs can
be found through a partner login. Sadly, the only public docs I've found
so far are the exact same as you have just sent.
The latter of which is an old bug we used to suffer frequently on our
Citrix mainframes. (Ah, the reason he's been screwing with process and
memory management!).
N.B. Did ya miss MS's other two support sites? (just joking, I _love_
the way they make us trawl 5 different places for info ;-).
--
None of these negate what I have said. In fact all they do is negate
what you said about never changing from defaults, as here you will find
MS suggesting that people look at these articles for their solutions
(having been on the phone for hours to MS waiting for some id-10t (it's
been a while) to point me to Q184419, despite being outdated and useless
to the problems we ever encounter on the citrix platforms. The techs are
never as good as the consultants. :-(
Back to the point in hand, setting this option can reduce paging in
certain instances where the data paged is stored in driver space - this
is particularly more common in certian development scnearios. You CAN
analyse these effects by careful observation of kernel and process
running times, along with changes in memory deltas and other less
important varaibles. Generally you will need a program in ring 0 to
properly observe these things though, and never forget that everything
you do on the system affects the system.
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives, please
visit:
http://list.valvesoftware.com/mailman/listinfo/hlds
--
Clayton Macleod
get ye flask
You cannot get ye flask.
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives, please
visit:
http://list.valvesoftware.com/mailman/listinfo/hlds
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives, please
visit:
http://list.valvesoftware.com/mailman/listinfo/hlds
